README.md

Heat engine gym

A collection of heat engines, based on the OpenAI Gym environment framework for use with reinforcement learning applications. If you use this gym, please cite

@article{beeler2021optimizing,
  title = {Optimizing thermodynamic trajectories using evolutionary and gradient-based reinforcement learning},
  author = {Beeler, Chris and Yahorau, Uladzimir and Coles, Rory and Mills, Kyle and Whitelam, Stephen and Tamblyn, Isaac},
  journal = {Phys. Rev. E},
  volume = {104},
  issue = {6},
  pages = {064128},
  numpages = {10},
  year = {2021},
  month = {Dec},
  publisher = {American Physical Society},
  doi = {10.1103/PhysRevE.104.064128},
  url = {https://link.aps.org/doi/10.1103/PhysRevE.104.064128}
}

To install:

pip install ./

Environments

Environment	Description
Carnot-v0	Heat engine environment with action space restricted to the required actions.
Carnot-v1	Heat engine environment with full action space.
Carnot-v2	Heat engine environment with full action space and discrete variable dV actions.
Carnot-v3	Heat engine environment with full action space and continuous variable dV actions.
Carnot-v4	Heat engine environment with full action space, W and Qin included in the state, and finite W and Qin available. Change in W after final step is used for reward.
Stirling-v0	Heat engine environment with adiabatic actions unavailable.
Stirling-v1	Heat engine environment with adiabatic actions unavailable and variable dV actions.
Otto-v0	Heat engine environment with isothermal actions unavailable.
Otto-v1	Heat engine environment with isothermal actions unavailable and variable dV actions.
Beeler-v0	Heat engine environment with adiabatic actions replaced with irreversible actions.
Beeler-v1	Heat engine environment with adiabatic actions replaced with irreversible actions and variable dV actions.

Version history:

• Version 1.0 released 2018-11-11.

To run

See demo.py for an example.